Question

我使用Flask制作了一个API，该API使用预先训练的Scikit学习模型提供了预测。该API在本地主机和Heroku本地网络上都可以正常运行，但是在部署模型时无法加载该模型。 API以JSON形式返回预测。在本地运行时返回{"prediction":123}。部署后将返回{"error":"Failed to load model"}。

应用架构：

 - houseprediction.py
 - waitressServer.py
 - requirements.txt
 - Procfile
 - supp-files:
    - gbModel.pkl 
    - model_columns.pkl
    - sectorLabels.pkl

houseprediction.py：

from flask import Flask, jsonify, request
from sklearn.externals import joblib
import pandas as pd
import os

app = Flask(__name__)
@app.route('/predict', methods=['POST'])

def predict():
    dir_path = os.path.dirname(os.path.realpath(__file__))
    try:
        gbModel = joblib.load("{}\\supp-files\\gbModel.pkl".format(dir_path))
    except:
        return jsonify({'error': 'Failed to load model'})
    try:
        model_columns = joblib.load("{}\\supp-files\\model_columns.pkl".format(dir_path))
    except:
        return jsonify({'error': 'Failed to load model columns'})
    try:
        lbl = joblib.load("{}\\supp-files\\sectorLabels.pkl".format(dir_path))
    except:
        return jsonify({'error': 'Failed to load sector labels'})
    json_ = request.get_json()
    query_df = pd.DataFrame(json_, index=[0])
    pd.options.display.max_columns = 50
    query_df['sector'] = lbl.transform([query_df['sector']])[0]
    print(query_df.dtypes)
    query = pd.get_dummies(query_df)
    for col in model_columns:
         if col not in query.columns:
              query[col] = 0
    print(query_df.shape)
    print(query_df.head(1))
    prediction = gbModel.predict(query)
    print(prediction)
    return jsonify({'prediction': prediction[0]})

@app.route('/')
def home():
    return "Welcome to House Prediction"

waitress.py：

from waitress import serve
import os
import housepredictionServer     
serve(housepredictionServer.app, port=os.environ['PORT'])

Procfile：web: python waitressServer.py

让我知道是否还有其他需要。

Answer 1

将所有路径更改为library(MASS) library(dplyr) #> #> Attaching package: 'dplyr' #> The following object is masked from 'package:MASS': #> #> select #> The following objects are masked from 'package:stats': #> #> filter, lag #> The following objects are masked from 'package:base': #> #> intersect, setdiff, setequal, union data("iris") iris$Species<-NULL C <- cov(iris) # Note covariance, not cor mu <- colMeans(iris) df <- as_tibble(mvrnorm(10000,mu,C)) cor(df) #> Sepal.Length Sepal.Width Petal.Length Petal.Width #> Sepal.Length 1.0000000 -0.1269997 0.8728482 0.8220813 #> Sepal.Width -0.1269997 1.0000000 -0.4334930 -0.3707066 #> Petal.Length 0.8728482 -0.4334930 1.0000000 0.9633552 #> Petal.Width 0.8220813 -0.3707066 0.9633552 1.0000000 cor(iris) #> Sepal.Length Sepal.Width Petal.Length Petal.Width #> Sepal.Length 1.0000000 -0.1175698 0.8717538 0.8179411 #> Sepal.Width -0.1175698 1.0000000 -0.4284401 -0.3661259 #> Petal.Length 0.8717538 -0.4284401 1.0000000 0.9628654 #> Petal.Width 0.8179411 -0.3661259 0.9628654 1.0000000，因为heroku是基于Unix的系统

在Heroku上部署后，文件未加载。在localhost和heroku本地网络上运行良好

1 个答案: